[AliAbouelazm]

Fixes #15726.

Schema inference resets the full DataFrame index before selecting values, which raises when an index and column share a name. Select the column or named index level directly instead, preserving the documented column precedence and resetting only the selected Series labels.

This fixes both the BigQuery fallback and pandas-gbq inference. Regression tests cover object/numeric values, duplicate row labels, mixed MultiIndexes, null/empty data, schema overrides, index exclusion, and unchanged input.

Validation:

• Full unit suites on Python 3.12: BigQuery 2,949 passed, 2 skipped; pandas-gbq 268 passed.
• Focused helper/schema tests on Python 3.10 through 3.15, including pandas 1.5.3, 2.3.3, and 3.x.
• Both package linters and mypy pass. All changed executable lines and branches are covered, with no package coverage decrease.
• Offline integration checks pass. Cloud system tests were not run.

The delegated BigQuery path needs a pandas-gbq release containing this fix. No dependency bump is included here.

AI assistance was used for implementation, testing, and review.

[gemini-code-assist]

Code Review

This pull request updates the schema inference logic for pandas DataFrames in both google-cloud-bigquery and pandas-gbq to handle cases where an index shares a name with a column. Instead of resetting the index, columns and index levels are now retrieved directly, with columns taking precedence. However, a critical bug was identified in _pandas_helpers.py where the variable values is defined conditionally inside an if block but used unconditionally later, which would lead to an UnboundLocalError if the condition is not met.

[gemini-code-assist]

If bq_type is not None, the if bq_type is None: block is skipped, which means values is never defined. When the execution falls through to Step 3, calling _get_schema_by_pyarrow(column, values) will raise an UnboundLocalError. To fix this, define values unconditionally before checking bq_type.

Suggested change

	bq_type = _PANDAS_DTYPE_TO_BQ.get(dtype.name)
	if bq_type is None:
	sample_data = _first_valid(dataframe_reset_index[column])
	values = get_column_or_index(dataframe, column)
	sample_data = _first_valid(values)
	values = get_column_or_index(dataframe, column)
	bq_type = _PANDAS_DTYPE_TO_BQ.get(dtype.name)
	if bq_type is None:
	sample_data = _first_valid(values)